The week of June 1 to 7, 2026 made one thing clear: the AI industry’s bills are coming due. Anthropic raced to file for an IPO, developers revolted over metered pricing, and the power grid started buckling under the weight of new data centers. Here are the ten biggest AI stories of the week, counting down to the one that mattered most.

10. Physical AI keeps raising money

Robotics startups pulled in serious cash this week, led by Generalist’s $400M round to scale general-purpose robot models it says lift task success from 64% to 99%. Nvidia open-sourced Cosmos 3, an open model for physical reasoning and action, while startups from Hello Robot to Genesis AI pushed home assistants and faster simulation. The shared bet across all of them is that better training data, not new hardware, is what robotics needs next.

9. Local-first AI lands on your laptop

Google released Gemma 4 12B, an open-weights model under a permissive Apache 2.0 license that runs entirely on a typical 16GB laptop and still handles audio and video. The move doubles down on small on-device models while rivals chase ever-larger ones. It arrived alongside hardware aimed at the same goal, including Microsoft’s Surface RTX Spark Dev Box for running big models without cloud bills.

8. OpenAI turns Codex into an everyday work platform

OpenAI reframed Codex from a coding tool into an operating environment for knowledge work, shipping role-specific plugins for analysts, marketers, designers and bankers, plus web hosting and inline annotations. Its models and Codex also went generally available on Amazon Bedrock. One senior employee went further, telling TechCrunch that “Chat is dead” as the company works on a super app.

7. Nvidia becomes a PC chipmaker

At Computex, Nvidia unveiled the RTX Spark, its first complete Arm-based CPU-plus-GPU chip for laptops and mini-PCs, calling it the most efficient PC chip ever built. Microsoft signed on as launch partner, putting the chip inside a new Surface Laptop Ultra and reviving a bet it once wrote off. Overnight, Nvidia became a direct rival to Intel, AMD, Apple and Qualcomm in consumer computing.

6. AI policy splinters along national lines

President Trump signed a watered-down executive order that only asks labs to voluntarily share frontier models before release, after industry pushback killed a tougher draft. The retreat coincided with his White House AI advisor stepping down and reports the government might take an equity stake in OpenAI. Allies are heading the other way, with Canada joining a growing push to build sovereign AI and cut reliance on US systems.

5. The compute buildout collides with the power grid

The money pouring into infrastructure reached staggering numbers, with Google agreeing to pay SpaceX $920M a month for compute and SoftBank pledging up to 75 billion euros for French data centers. But the physical limits are showing: Texas grid operators flagged reliability risks after data centers failed voltage tests, and New York advanced a one-year moratorium on new builds. The compute story is quietly becoming a power story.

4. Microsoft breaks free from OpenAI

At Build 2026, Microsoft unveiled seven in-house MAI models, led by a trillion-parameter reasoning model, and its AI chief said the company had been “set free” from OpenAI to chase superintelligence. It also launched Scout, an always-on assistant inside Microsoft 365, and Project Solara, an Android variant built for agents instead of apps. After years defined by its OpenAI partnership, Microsoft is writing its own story.

3. AI agents become the hacker’s new weapon

This was the week agentic AI’s security bill came due. Hackers simply asked Meta’s support chatbot to hand over high-profile Instagram accounts and it worked, researchers turned OpenAI’s Codex into an HTTP/2 Bomb that crashes servers in seconds, and a single malicious GitHub issue was enough to hijack a Claude Code agent. OpenAI responded with a Lockdown Mode for ChatGPT, while admitting it only reduces the odds of a leak rather than closing the hole.

2. The cost of AI starts to bite

The era of cheap, subsidized inference looks to be ending, and users noticed fast. GitHub Copilot’s switch to usage-based pricing triggered a revolt, with some developers burning a month of credits in a single day, while Uber capped employee AI spending after blowing its budget in four months. TechCrunch dubbed the wave of rising token prices the “Tokenpocalypse”, and even AI boosters began asking whether the subscription is still worth it.

1. Anthropic files for a record IPO, with Claude writing its code

The maker of Claude confidentially filed paperwork for what analysts say could be the largest IPO ever, weeks after overtaking OpenAI’s valuation and reporting annualized revenue above $47 billion in May, up from roughly $9 billion at the end of 2025. In the same breath, Anthropic revealed that more than 80% of the code merged into its production codebase that month was written by Claude, framing it as early progress toward recursive self-improvement. The company then argued the industry should slow down its “AI sprints”, a striking message from the firm sprinting hardest of all.

If one theme tied the week together, it was gravity. The AI industry spent two years defying it, and the bills, the grid, and the regulators all arrived at once.